SB v2.4 released

[jjjjL]

Just wanted to let everyone know about the new client.

It's a little faster and solves some other issues.

Head over to http://www.seventeenorbust.com/ to upgrade.

Cheers,
Louie

[Matt]

Once again the FreeBSD client is neglected :( :swear:

If you need a box to compile on I can let you have access.

[vjs]

Great!!!


Windows client can accept longer usernames

Looks like you can now do secondpass without the registry hack

usernameQQQsecondpass

works!!!! :bouncy: :elephant: :bouncy:

[Alien88]

Originally posted by Matt
Once again the FreeBSD client is neglected :( :swear:

If you need a box to compile on I can let you have access.

fbsd client is updated.

[hhh]

Beerknurd wrote somewhere else:

The new version is slow as crap on my P4 3.4 Ghz machine.... Will tests still be accepted using version 2.3???
Just for info. H.

[jjjjL]

Originally posted by hhh
Beerknurd wrote somewhere else:

The new version is slow as crap on my P4 3.4 Ghz machine.... Will tests still be accepted using version 2.3???

I would discourage the use of v2.3, especially on P4s. If the v2.4 client slowed down a test you were doing, that means it was one of the boundry cases where the FFT was too low before and the new client chose a more appropriate (slower) FFT. I'm glad you benchmarked it to try and optimize your system to produce as much as possible, but believe me when I say it would be doing a huge dis-service to the SB community for people to run v2.3 instead of v2.4 to increase the speed on a few tests. v2.4 is actually faster on every processor and OS for every size FFT so for the majority of tests, it would make sense to use v2.4 even if you were only concerned with pure stats.

The upgrade is most critical for P4 Linux users but there will be a speed increase for others too so everyone should consider upgrading.

Cheers,
Louie

[kelman66]

Do I have to wait for my current work unit to finish?

[Beerknurd]

I just realized I have been using v2.0 SSE2..... Is that bad??? :gangpunch :D

They are now all updated..... oops!!! :D

By the way... It is faster now.....

:notworthy Version 2.4

[kelman66]

Originally posted by kelman66
Do I have to wait for my current work unit to finish?

To answer my own question: No.

and it 2.4 is definitely faster on my P4 2.8Ghz machine

[Polski Radon]

I've noticed that this version is faster when you double the instances running.

I went from 2.8 to 3.0M cEM's/second by running 4 instances.

[kelman66]

Interesting, I was going to ask this question yesterday.

I can run more than one instance on the same machine?

With my P4, system Idle Process is always taking 50% and I wondered if I could open another one and use some of that idle time.

[Frodo42]

Originally posted by kelman66
Interesting, I was going to ask this question yesterday.

I can run more than one instance on the same machine?

With my P4, system Idle Process is always taking 50% and I wondered if I could open another one and use some of that idle time.
As I understand it, System Idle process is hyper threading ... those 50 % do go to SB.
But I do think that there may be a small gain by running more instances of the process ... the best way to figure it out is to do some benchmarking ... it will probably be diffrent from system to system. I am don't know enough about all this L1 and L2 cache stuff etc. to be able to tell you why things act the way they do.

Anyhow just be careful not to get to long completion time ... what helps the projects is the tests you complete not how many cems/s you produce.

Also it could be interesting to see if running more processes gives higher error-rates ... but that would require either that one of the guys with acces to error rates check it (I don't think there is a big chance for that as both Louie and David seem to be pretty overloaded with things to do right now) or better yet, that some data on error rates are added to the stat's pages.

[kelman66]

ya i knew about the hyperthreading thing and this is why I havent done it yet.
But I do run prime 95 simultaneously with it sometimes and I find that it is slightly more productive because each runs slightly more than 1/2 as fast (say 60%).

So instead of 0.5 + 0.5, it is more like 0.6 + 0.5 or 0.6 + 0.6 and I can get 10-20% more work done over the course of a week. SB seems less affected by running prime95 than prime95 is by SB.

So IF the same thing applies when two instances of SB are running, I should get 10-20% more results produced over the same course of time.

Seems like this meets BOTH desires to increase cems but ALSO produce more results. If it takes twice as long but you produce 120% more, you are ahead.
Even if it was only 5%, that would be a lot system-wide.

As mentioned, error rates would be important to check out.

[Mooch12345]

How do you run more than one instance at a time on a machine? The only thing I've managed to accomplish is running the same test in multiple windows.

[Polski Radon]

Using WinXP?
I hereby state that the arrow '>' means press/hit/simulate the Enter key.

Do: Start > Run > cmd

Then: cd.. > cd.. > cd sb
(If your directory is C:\sb. Otherwise substitute to your installed directory)


Then: sobsvc -i
(To Install the SoB service. This will show up in the Services. You can see the Services by doing Start > Run > services.msc )


To install 2 services (2 instances) with aggressive use of Hyperthreading
Do: sobsvc -o2


To install 4 services, like I did.
Do: sobsvc -p:4:1


I would also reccomend you set the following flags:
sobsvc -m
sobsvc -s
sobsvc -k

[mgl]

For anyone interested, the 2.4.0 Linux client to successfully run under OpenBSD 3.5, 3.6 and 3.7 linux emulation. I'll put my hand up to port it, as it core dumps on dual proc machines, but otherwise I'll persist with it as it is.

[GP500]

Anyone with a pentium 1.6 M (centrino)

what is your speed cems/sec.

[stim]

Hey all,

Just wondering if there will be an update to the BeOS client?

Thanks.

[Greenbank]

And any chance of static linux binaries?

I've got an old version of linux on a P4 1.7GHz that I'm having to sieve on because sbclient won't work and I can't upgrade libc on that machine.

[Alien88]

Originally posted by Greenbank
And any chance of static linux binaries?

I've got an old version of linux on a P4 1.7GHz that I'm having to sieve on because sbclient won't work and I can't upgrade libc on that machine.

I thought they were static.... what problem do you have running the client?

Feel free to e-mail me as it will be faster - mgarrison @ alienz.net

[Greenbank]

# file sb
sb: ELF 32-bit LSB executable, Intel 80386, version 1, dynamically linked (uses shared libs), not stripped

# ldd sb
./sb: /lib/i686/libc.so.6: version `GLIBC_2.3' not found (required by ./sb)
libm.so.6 => /lib/i686/libm.so.6 (0x4002e000)
libstdc++.so.5 => not found
libc.so.6 => /lib/i686/libc.so.6 (0x40052000)
/lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)

I only have libc-2.2.2.so on this machine.

I can't upgrade anything as it runs some legacy software that doesn't work with libc-2.3.x or above.

If sb was compiled static then I'd be able to use this P4 for prp-ing instead of (slow) sieving.

[Theadalus]

Hmmm, i noticed (with the Windows version) the timer problem still isn't solved: after a while "time left" is running synchronized with "run time", exact same values! :weggy:

[Bok]

Originally posted by Greenbank
# file sb
sb: ELF 32-bit LSB executable, Intel 80386, version 1, dynamically linked (uses shared libs), not stripped

# ldd sb
./sb: /lib/i686/libc.so.6: version `GLIBC_2.3' not found (required by ./sb)
libm.so.6 => /lib/i686/libm.so.6 (0x4002e000)
libstdc++.so.5 => not found
libc.so.6 => /lib/i686/libc.so.6 (0x40052000)
/lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)

I only have libc-2.2.2.so on this machine.

I can't upgrade anything as it runs some legacy software that doesn't work with libc-2.3.x or above.

If sb was compiled static then I'd be able to use this P4 for prp-ing instead of (slow) sieving.

I guess you could try doing a symlink from whatever version of libstdc++.so.xxx you have to libstdc++.so.5.

Depends on which functions they are using within it, but it has a chance.

Bok

[Greenbank]

Tried that, but then libc moans about various versions of other libraries, there's no easy way for me to fix it.

I'll put it another way. I have access to ~10 similar machines at work, all this version of libc, that could run sbclient, but it's not possible to run it.

A static compiled binary would be the simplest solution.